Optimizing Classification-Based Machine Learning Models for Bacterial Spot Disease Detection in Tomato Plants

Plant pests and diseases destroy up to 40% of global crop production each year, causing more than US$220 billion in losses [4]. Tomatoes are among the most widely consumed crops, with approximately 160 million tons consumed each year [13]. In Florida, bacterial spot, caused mainly by Xanthomonas perforans [9, 11] and favored by hot, humid, rainy weather, cost tomato growers almost $900,000 in the 2007–2008 season [2]. We hypothesized that infected tomato leaves differ measurably from healthy leaves in color and texture, and tested whether machine learning could detect these differences. Using 3,769 PlantVillage leaf images, we compared Logistic Regression, Ridge Classifier, Decision Tree, Random Forest, and Support Vector Classifier (SVC) models with a lightweight convolutional neural network (CNN). The tuned SVC performed best among the classical models (99.34% accuracy, 99.44% F1, ROC-AUC 1.00) and matched the CNN (McNemar’s test, p = 1.00). However, its accuracy fell sharply on rotated or brightened images, so field performance is likely lower. Machine learning shows promise for early detection of bacterial spot, but it must be validated on field images before practical use.